The name COBYLA is an acronym for constrained optimization by linear approximation. It is an iterative method for derivative-free constrained optimization. The method maintains and updates linear approximations to both the objective function and to each constraint. The approximations are based on objective and constraint values computed at the vertices of a well-formed simplex. Each iteration solves a linear programming problem inside a
trust region whose radius decreases as the method progresses toward a constrained optimum.